| Age | Commit message (Collapse) | Author |
|---|
|
Each app now gets two ZFS datasets - a disposable OS layer
(zroot/skiff/appbase/<app>) cloned from base, and a persistent data
layer (zroot/skiff/appdata/<app>) holding releases/current/env/
meta.json/data.db. `skiff rebuild --app|--all` discards and reclones
the OS layer to pick up an OS/toolchain/config-template upgrade,
guarded by the same snapshot-and-rollback safety net as deploy and
config changes; the data layer is never touched.
Also fixes stale docs (builds already run via jexec inside the jail,
not on the host; --iface already auto-aliases the jail's IP, so the
manual ifconfig step and its troubleshooting entry are gone) found
while updating the same sections for this feature.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
Builds previously ran as a bare host process, sharing one host-wide
toolchain across every app - a Go/gcc upgrade on the host silently
changed what the next deploy of any app picked up, and cgo builds
needed the exact right headers/compiler on the host regardless of
what each app actually targets. Routing the build through jexec, the
same mechanism release.Run already uses for the release phase, gives
each app's own persistent jail an independently-managed toolchain
instead.
|
|
- Extract deploy.Run's snapshot -> apply -> restart -> health-check ->
rollback-on-failure sequence into deploy.GuardedApply, shared with
cmd/skiff's applyConfigChange, which duplicated it verbatim.
- Clean up the worktree and snapshot a deploy creates when it fails
before cutover (e.g. a failing release-phase migration), so retrying
the same sha no longer needs manual cleanup.
- Make checkout and snapshotting idempotent for a same-sha retry after
a cutover that later got rolled back on a failed health check, where
the release dir and snapshot are deliberately left behind by design.
- Document the actual cause of a jail not surviving a host reboot:
FreeBSD ships no /etc/jail.conf, so boot-time `service jail start`
finds nothing to start regardless of jail.conf.d - only name-targeted
starts read jail.conf.d directly, which is why manual starts worked.
|
|
Config vars are plain env vars, injected into both the running app (via
rc.d) and the release phase, and versioned on the same ZFS snapshot
timeline as deploys - a bad config change gets the same automatic
snapshot/restart/health-check/rollback guarantee a bad deploy already
gets. Extracts deploy.go's health check into internal/health so both
paths share it.
Verified end-to-end against a FreeBSD VM: config set/unset, automatic
rollback on a health-check failure, and explicit `skiff rollback` all
correctly carry the env file along with the rest of the dataset.
|
|
|
